-
Ragin’ Cajun RoboBoat 1
Ragin’ Cajuns RoboBoat–2020
Captains:Benjamin Armentor and Joseph Stevens
Members:Gerald Eaglin, Bradley Este, Thomas Poché,
Nathan Madsen, Dallas Mitchell, Andrew DurandCoach:
Joshua Vaughan1
Abstract— This report discusses the motivations behindthe design
choices and improvements to the Universityof Louisiana at
Lafayette’s first entry to RoboNation’sRoboBoat Competition. Due to
restrictions on groupgatherings and university resources, this
design has notbeen physically constructed or tested. The Ragin’
CajunsRoboBoat is a catamaran-style vessel equipped with
fourthrusters in an “X”-Configuration, enabling holonomicmotion.
The computer network communicates with in-dividual components via
the Robot Operating System(ROS). The main contributions from the
2020 Ragin’Cajuns RoboBoat team include a new control
method,procurement of a larger electronics enclosure, upgradingthe
computer vision hardware, and creating a simulationof the system in
Gazebo.
I. INTRODUCTIONThe 2020 RoboBoat competition requires teams
to build an Autonomous Surface Vessel (ASV)capable of performing
a variety of tasks. For anASV to accomplish these tasks, several
subsys-tems must function together. The University ofLouisiana at
Lafayette has developed a designto compete in the 2020 RoboBoat
competition,shown in Figure 1. The ASV is equipped withtwo 2D LiDAR
rangefinders and two stereo cam-eras for vision feedback, a GPS and
IMU forlocalization, and four thrusters mounted in
an“X”–Configuration, enabling holonomic motion.Because the
in-person portion of this year’s com-petition was cancelled, the
design process focused
1Department of Mechanical Engineering, Universityof Louisiana at
Lafayette, Lafayette, LA 70504, [email protected]
Fig. 1: 2020 Ragin’ Cajuns RoboBoat CADModel
on upgrading and developing new tools used totest future
designs.
II. COMPETITION STRATEGY
This section will discuss the 2020 Ragin’ Ca-jun RoboBoat team’s
approach to completing thetasks set out by RoboNation for this
year’s compe-tition. The following subsections are in the orderthat
the tasks would be attempted.
A. Navigation Channel
To demonstrate autonomy and ensure some levelof course safety,
this task is mandatory andmust be completed before any other tasks
canbe attempted. The ASV must pass through twosets of gates. Each
gate consists of two buoysat least six feet apart. The two sets of
buoys
-
Ragin’ Cajun RoboBoat 2
Fig. 2: State Machine Behaviors for Acoustic Docking Task
are at least 50 feet apart. The Ragin’ CajunsRoboBoat is
equipped with a stereoscopic camerathat provides images to an image
classifier trainedby a Convolution Neural Network (CNN).
Thetraining set for this CNN consists of manually-labeled images
from the previous competition, aswell as images from the 2016
Maritime RobotXCompetition. The output of the image classifieris
made available to a state machine that deter-mines how the ASV
should maneuver. For theNavigation Channel task, the state machine
directsthe ASV to find a gate, identified with green inthe right
side of the frame and red in the left. Awaypoint goal is sent to
the navigation stack tomaneuver to the middle of the gate and
orient thevessel to be in-line with the channel. The statemachine
instructs the ASV to maintain the initialheading as it continues to
drive forward and lookfor the exit gate. Another waypoint between
theexit gate is sent to the navigation stack as a targetlocation.
As the vessel reaches this waypoint, itcompletes the Navigation
Channel and proceedsto the next task.
B. Acoustic DockingThis task requires the ASV to localize a
signal
and dock in its location. The state machine be-haviors for this
task are shown in Figure 2. Oncethe Navigation Channel task has
been successfullycompleted, the state machine instructs the ASV
to
maneuver to the GPS coordinate provided for theentrance to the
Acoustic Docking task, avoidingpotential obstacles along the way.
Once the ASVarrives at the docking station, two hydrophonesare
deployed from the vessel’s stern using a linearactuator. The ASV
then circumnavigates the dockto generate a map of the area.
Hydrophone feed-back is processed to localize the active
acousticbeacon and record its location. Once the signal hasbeen
located, if the docking station is not in view,the state machine
will prescribe a waypoint awayfrom the dock and instruct the ASV to
adjust itsheading to see the circle, cruciform, or trianglesymbol
associated with the recorded location.This symbol is identified
using the image classifierand recorded, as it may affect the tasks
that follow.The recorded location of the active acoustic bea-con is
then used as a waypoint to maneuver backto the dock. When the ASV
has reached the targetlocation, docking will commence. The ASV
willstation keep in the dock for five to ten seconds toconfirm that
it has successfully docked, and thenexit the dock by setting the
GPS coordinates forthe Obstacle Channel task as its next
waypoint.
C. Obstacle ChannelThe obstacle channel requires the ASV to
ma-
neuver through a series of gates with obstaclebuoys along the
trajectory. Unlike the Naviga-tion Channel task, these gates are
arranged in a
-
Ragin’ Cajun RoboBoat 3
Fig. 3: State Machine Behaviors for Obstacle Field Task
non-linear trajectory. Because the Ragin’ CajunRoboBoat hull
footprint is large, this task may bedifficult to complete. However,
our ASV is alsoequipped with a holonomic thruster
configuration.This allows the ASV to exert forces and momentsin
each degree of freedom independently. Thisincreased
maneuverability, along with restrictionson maximum allowable
velocities, should aid theRagin’ Cajun RoboBoat to complete this
task.
Once the ASV has arrived at this task from theAcoustic Docking
station, the state machine willbe checking for buoy color. It will
be looking forgreen buoys in the right side of the image and
redbuoys in the right to identify the gates. The visionfeedback
system provided to the navigation stackwill prevent the ASV from
colliding with thesegreen and red buoys as well as the yellow
obstaclebuoys. Waypoints are recursively placed at themiddle of the
gates. In order to know that the taskhas been completed, the GPS
coordinate for theObstacle Field task is iteratively appended to
thewaypoint sequence. When no more gates are inthe field of view,
the ASV will begin maneuveringto the Obstacle Field task because
this waypoint isthe last in the sequence. If red and green buoys
inthe Obstacle Field are misinterpreted as gates, thepath planner
will prevent the ASV from enteringa region that it cannot fit
because it knows thebase footprint area.
D. Obstacle Field
This task is attempted immediately after the Ob-stacle Channel
task. The state machine behaviorsfor this task are shown in Figure
3. The Obstacle
Field task is similar to the Obstacle Channel,but instead of
gates, there is one “Pill Buoy”,which has distinguished markings,
that the ASVmust circumnavigate. This buoy is surrounded byseveral
obstacles spaced as little as four feet apart.The Ragin’ Cajun
RoboBoat team’s approach tothis task is handled by collaboration
between thehigh-level state machine and the path planner.The state
machine will determine that the ASVis at the Obstacle Field by
first identifying thePill Buoy in the cameras’ field of view. The
statemachine will then place a waypoint toward thePill Buoy to
motivate the ASV to approach it.An entrance to the Obstacle Field
is found bycircumnavigating the field of buoys, iteratively
up-dating the largest distance identified between theadjacent
buoys. The path planner will then plan atrajectory through the
obstacles while maintaininga safe distance from obstacles that is
manuallyprescribed. This value is chosen as smaller than50% of the
difference between the minimumspecified distance between obstacles
the ASV canpass through, four feet, and the beam width ofthe Ragin’
Cajun RoboBoat. When the ASV hasentered the Obstacle Field the
current position isrecorded, the state machine will command thepath
planner to circumnavigate the obstacle byplacing waypoints around
it. Once the Pill Buoyhas been circled, and the ASV has changed
itsheading by at least 360◦, the ASV will exit theObstacle Field
using the recorded position as thelast waypoint and locating of a
pair of buoysthe computer vision system and path planner findwide
enough for the ASV to fit.
-
Ragin’ Cajun RoboBoat 4
E. Speed Gate
After the obstacle field has been successfullyescaped, the ASV
will maneuver to the GPScoordinate provided for the Speed Gate
entrance.This task introduces a race element to the com-petition.
The ASV is required to enter througha gate, travel to a buoy and
circle it, and returnthrough the gate as quickly as possible.
Becauseof the new enclosure that was procured for thiscompetition,
the ASV thrust-to-weight ratio hasdecreased, hurting our possible
performance atthis task. However, the new optimal control strat-egy
implemented for this year’s competition maybe more effective than
the velocity controller usedat the previous competition.
Additionally, the statemachine overrides the velocity restrictions
placedon the ASV by the path planner for this taskso it can be
completed as quickly as possible.A waypoint is placed at the gate
entrance, andthen the state machine instructs the ASV to
passthrough the gate at maximum speed. Once thetarget buoy has been
located, waypoints are placedaround it to it can be
circumnavigated. The initialposition at the gate is used as the
final waypoint toguide the ASV out of the Speed Gate Challenge.
F. Return to Dock
Once all other tasks have been completed, thefinal task the ASV
can complete is returning to thestarting point of the course
without encounteringany obstacles. The ASV must pass through itsown
course, and not another where other vesselsare competing. This is
accomplished by recordingthe starting position upon entering the
water,and then upon completing the final competitiontask,
activating that saved location as a waypoint.While the ASV is
completing other competitiontasks, it will also be generating a map
of theenvironment. In conjunction with its localizationand path
planning capabilities, this will be usedto maneuver the ASV back to
the starting dock.
G. Object Delivery
This task requires the ASV to deliver up tofour objects to a
specified area in the course. Thetask may be completed solely by
the ASV or by
Fig. 4: Free-Body Diagram of RoboBoat ThrusterConfiguration
a combination of an ASV or Unmanned AerialVehicle (UAV). The
2020 Ragin’ Cajun RoboBoatteam declined to develop this task to
focus onother hardware and software upgrades. Should atask like
this be introduced in future competitions,the Ragin’ Cajun RoboBoat
would have beentasked with collecting images of the new
courseelements to enhance the image classifier trainingset, which
currently has a limited supply of dock-like images.
III. DESIGN CREATIVITY
A. Thrust ConfigurationThe thrusters are configured in an “X”
pattern,
and mounted at 45◦ angles, relative to the bow. Afree-body
diagram of this thruster configurationis shown in Figure 4 This
enables holonomicmotion, or motion in all three degrees of
freedomindependent of each other. This improves the
ma-neuverability of the RoboBoat through tasks likethe Obstacle
Channel or Obstacle Field. It alsoassists in docking since the
RoboBoat can movein the positive or negative sway directions
anddoes not have to do a forward–reverse maneuver,such as a car
parallel parking.
Though this thrust configuration was used on the2019 Ragin’
Cajun entry to the competition, thedesign was only equipped with
one stereo cameraand LiDAR. This meant that there was a 90◦
-
Ragin’ Cajun RoboBoat 5
region that no visual measurements were beingtaken, so it was
possible that, with a mappingerror, the 2019 RoboBoat would apply
reverse orsway thrust and encounter an obstacle. Becausethe 2020
design has two LiDAR systems andstereo cameras, the blind spot is
removed.
B. Control StrategyThe 2020 Ragin’ Cajun RoboBoat control
sys-
tem has been upgraded from last year’s entry touse Model
Predictive Control (MPC). This is anoptimal control method that
solves for the optimalinput sequence over the next n time steps
giventhe current states of the system by minimizinga cost function.
The first input in the optimalcontrol sequence is supplied to the
system andthe process is repeated. Using this new optimalcontrol
strategy requires some system parametersto be estimated so the
equations of motion can besolved to predict trajectories.
Figure 5 illustrates this process for a singledegree of freedom
system tracking a constantsetpoint. In Figure 5a, the controller
calculates theoptimal control sequence and the system advancesto a
new state. For the next time step, shown inFigure 5b, new initial
conditions are provided tothe controller and the optimal control
sequenceis computed again. The trajectory and predictedstates have
also extended one step past theirpositions in Figure 5a.
The equations of motion describing the dynam-ics of the Ragin’
Cajun RoboBoat are [1], [2]:
MMMν̇νν +CCC(ννν)ννν +DDD(ννν)ννν = τττ + τττwind + τττwave
(1)
η̇ηη =[ẋ ẏ ψ̇
]T (2)ννν =
[u v r
]T (3)where MMM is an inertia matrix, ννν is a vector
ofbody-fixed velocities, CCC (ννν) is a Coriolis matrixand DDD(ννν)
is a hydrodynamic drag matrix. Theparameters of the CCC (ννν) and
DDD(ννν) matrices mustbe determined empirically.
The controller minimizes a quadratic cost func-tion given
by:
J (x,u) =n
∑0
(xxxT QQQxxx+uuuT RRRuuu
)(4)
...
(a) Time t
...
(b) Time t +1
Fig. 5: Model Predictive Control Process for aSingle Degree of
Freedom System
QQQ = Diag([
0.5 0.5 1.0 0.8 0.5 0.1])
(5)
RRR = Diag([
0.001 0.001 0.001 0.001])
(6)
xxx =[∫
(νννd −ννν)dt(νννd −ννν)
](7)
uuu =[Tpb Tps Tsb Tss
]T (8)where QQQ and RRR are diagonal matrices, xxx is thevector
of state error from the set of desiredstates, and uuu is the
control input vector. Furtherexperimental testing is required to
fully tune thecost function weights.
-
Ragin’ Cajun RoboBoat 6
C. Thermal Energy Management
This year’s team was fortunate enough to receivea generous
donation from Hammond Manufac-turing and AWC, Inc. The team was
unable toconfigure the new electronics enclosure, but it hasbeen
integrated into our CAD model of the vessel.
Because the electronics have been upgraded toinclude an
additional CPU, the team was con-cerned about the amount of heat
generated in theenclosure, which is does not allow airflow. The
oldenclosure was a black polypropylene case with ahalf-inch wall
thickness. The new enclosure is afiberglass case that increases the
enclosure volumeby 160%. Furthermore, the enclosure has
beenelevated and equipped with two aluminum finnedplates. This
allows more space for air flow anda larger external surface area to
convect excessheat, but at the cost of an additional 10 poundsof
weight. This does reduce the Ragin’ CajunRoboBoat’s maximum speed
and thrust-to-weightratio.
Though more weight is being added, prelim-inary calculations
show that given the averagetemperature conditions at the 2019
competition,this enclosure will convect approximately 51%more
energy off of the enclosure surface based onthe increased area.
This does not take the changein temperature resulting from a gray
surface intoconsideration.
IV. EXPERIMENTAL RESULTS
A. Pool Testing
To develop parts of the system model, the Ragin’Cajun RoboBoat
was taken to a university poolto perform system identification
trials, shown inFigure 6. These trials were performed using withthe
old enclosure, but by estimating the increaseddraft increase the
following parameters have beenidentified. These same system
identification trialshave been performed on another
catamaran-stylevessel [3], [4]. Because the hull properties
aresimilar, the hydrodynamic added mass terms andnonlinear drag
terms have been estimated usingtheir formulations. The hydrodynamic
drag andadded mass term estimations for the adjusted
Fig. 6: Ragin’ Cajun RoboBoat Performing Sys-tem ID Trials
weight with the new enclosure are shown in Ta-ble I. The values
follow the SNAME convention,meaning subscripts represent drag force
in thatdegree of freedom [5]. For the nonlinear dragterms, the
subscripts represent drag in the firstsubscript due to motion in
the second subscript’sdegree of freedom. The terms Xu and Xuu
cor-respond to linear and nonlinear drag in surge.Because the
vessel’s weight has increased by 20%with the added enclosure, these
parameters willneed to be recollected. The other parameters
areempirically defined, but may be adjusted if thewaterline length
or draft is different than the newestimate from the additional
enclosure weight.
B. System Simulation1) Gazebo: The biggest contribution of
this
team to the continued development of the Ragin’Cajun RoboBoat is
the foundation for a systemsimulation in Gazebo. The RoboBoat
spawns inthe SandIsland World from the Virtual MaritimeRobotX
Competition [6], shown in Figure 7. Here,the image classifier
training is being tested on itsability to recognize the buoys. The
classifier uses“You Only Look Once” (YOLOv3) [7], trainedusing the
CNN described in Section II-A. Thesimulation is based on the Heron
Simulation fromClearPath Robotics [8]. The ROS network on-board the
RoboBoat computer systems and sen-sors has been migrated to
generic, configurablesensors with open-source plugins to simulate
data
-
Ragin’ Cajun RoboBoat 7
TABLE I: System Parameters
Item ValueMass 32.042 kg
Moment of Inertia (Z) 2.987 kgm2
Draft 0.1214 mLoa 1.524 mLwl 1.2827 mLcg 0.3191 mBoa 0.8128 mB
0.5588 mBh 0.2351 mXu 55.5 kg/sXuu 4.1 kg/sYv See [3], [4]Yr See
[3], [4]Nv See [3], [4]Nr See [3], [4]Yvv 171.292 kg/sYvr 55.190
kgm/sYrv 55.190 kg/sYrr 41.268 kgm/sNvv 55.190 kg/sNvr 41.268
kgm/sNrv 41.268 kg/sNrr 29.123 kgm/sXu̇ −1.602 kgYv̇ −53.451 kgYṙ
−11.926 kgmNv̇ −11.926 kgNṙ −17.713 kgm
such as point clouds, laser scans, and wind andwave effects.
The ACADO toolkit is used to generate a MPCcontroller [9].
Several other nodes are used toperform localization or act as
components in theROS Navigation Stack. The current
configurationstill needs lots of tuning, but the 2020 Ragin’Cajun
RoboBoat team decided that focusing ondeveloping this platform as a
tool for future com-petitions would be the best course of
action.
2) Computational Fluid Dynamics: With acompleted CAD model, the
team would like toverify the hand calculations for the
convectionheat transfer from the enclosure surface and per-form
thermal analyses within the enclosure itself.Though it is not ready
for this year’s competition,the updated CAD model will be used to
gener-ate these system characteristics before the 2021competition,
and may result in further hardwareupgrades such as an air
circulation system to be
Fig. 7: Testing the RoboBoat Image Classifier inSandisland
included within the enclosure.
V. CONCLUSIONS
This report analyzed the design of the 2020Ragin’ Cajun
RoboBoat, including key software,hardware, and strategic
improvements. This de-sign builds on the Ragin’ Cajuns’ previous
en-try to the RoboBoat Challenge, and now uti-lizes Model
Predictive Control, an optimal con-trol strategy. The sensor
configuration that theRoboBoat is equipped with now eliminates
ablind spot that hindered the holonomic thrusterconfiguration.
Contributions made by this teamto furthering ASV development at the
Universityof Louisiana at Lafayette include a Gazebo sim-ulation of
the RoboBoat system and an updatedCAD model. Competition strategies
employed anddocumented by this team may serve useful tofuture
Ragin’ Cajun RoboBoat team members.
VI. ACKNOWLEDGMENTS
This project would not be possible without theguidance and
support of our Coach, Dr. JoshuaVaughan. The team would like to
thank him forhis patience with us and willingness to assistus
wherever possible. The 2020 Ragin’ CajunRoboBoat team would also
like to thank Ham-mond Manufacturing and AWC, Inc. for their
-
Ragin’ Cajun RoboBoat 8
generous donation of a new and improved elec-tronics enclosure.
The team was unable to install itbecause of restricted access to
university facilities,but it will make its debut in the 2021 Ragin’
CajunRoboBoat entry.
REFERENCES[1] T. I. Fossen, Guidance and Control of Ocean
Vehicles. John
Wiley and Sons, Ltd., 1994.[2] ——, Handbook of Marine Craft
Hydrodynamics and Motion
Control. John Wiley and Sons, Ltd., April 2011.[3] W. B.
Klinger, I. R. Bertaska, K. D. von Ellenrieder, and
M. R. Dhanak, “Control of an unmanned surface vehicle
withuncertain discplacement and drag,” IEEE Journal of
OceanicEngineering, vol. 42, pp. 458–476, April 2017.
[4] E. I. Sarda, H. Qu, I. R. Bertaska, and K. D. von
Ellenrieder,“Station-keeping control of an unmanned surface vehicle
ex-posed to current and wind disturbances,” Ocean Engineering,vol.
127, pp. 305–324, November 2016.
[5] SNAME, “Nomenclature for treating the motion of a sub-merged
body through a fluid,” The Society of Naval Architectsand Marine
Engineers, Tech. Rep., 1950.
[6] B. Bingham, C. Agüero, M. McCarrin, J. Klamo, J. F.
Malia,K. Allen, T. Lum, M. Rawson, and R. Waqar, “Towardmaritime
robotic simulation in gazebo,” in OCEANS 2019,2019.
[7] J. Redmon and A. Farhadi, “Yolov3: An incremental
improve-ment,” University of Washington, Tech. Rep., 2018.
[8] C. Bogdon. Heron usv gets a new simulator.[9] B. Houska, H.
J. Ferreau, and M. Diehl, “Acado toolkit – an
open source framework for automatic control and
dynamicoptimization,” Optimal Control Applications and Methods,vol.
32, no. 3, pp. 298–312, 2011.
-
Ragin’ Cajun RoboBoat 9
APPENDIX
TABLE II: Ragin’ Cajun RoboBoat Specifications
Category Item Vendor Specifications Quantity Price ($)
Actuation HDA4-2 ServoCity 4” Stroke25lb Thrust 1 129.99
Battery 4S Li-Po Turnigy16V
5200 mAh450g
4 53.96
Battery 3S Li-Po Floureon12V
4500 maH324.5g
2 33.29
Comm. TL-WA901ND TP-Link
2.4-2.4835 GHz270m range
12V, 1A5.8W
1 37.99
Computing Pi 3B+ Raspberry Pi ARMv8, 1.4 Ghz1GB DDR2 RAM 2
35.00
Computing Jetson TX2 NVIDIA
256 CUDA Cores2-Core Denver 2
4-Core Cortex-A578GB DDR4 RAM
2 629.99
Enclosure PJ24208RT HammondMFG
0.064 m3
Fiberglass11 kg
1 Donated
Hull FiberglassCloth TotalBoat 6oz
yard210.56 yard2 56.01
Hull Epoxy TotalBoat 1.18 gcm3 4.31 kg 126.99
Hull FairingCompound TotalBoat 1.32g
cm3 2.27 kg 56.99
Propulsion T-200 BlueRobotics
[−4.1,5.25] kgf76mm Propeller156g (in water)
390W, 24A (max)
4 169.00
-
Ragin’ Cajun RoboBoat 10
Propulsion SpeedControllersBlue
Robotics
16.3g7–26V
30A (max)[1100,1900] µs
4 25.00
Sensing H2CHydrophoneAquarian
Audio
(0.01,100) KHz0.3mA
2KΩ ImpedanceOmnidirectional25mm x 58mm
51g≤80 meters
2 169.00
Sensing Scarlet 2i2 Focusrite (0.02,20)KHz1.5MΩ 1 159.99
Sensing UM6 IMU CH Robotics
500 Hz≤ 2◦ Pitch, Roll
≤ 5◦ Yaw5V
±2000◦/s rotation±2g accel.
1 1260.00
Sensing Ultimate GPSBreakout V3 Adafruit66 Channels
10 Hz5V, 20mA
1 39.95
Vision UTM-30-LX-EW Hokuyo
270◦ FOV2D Projection
30 meter range100 Hz
2 4900.00
Vision ZEDStereo Camera Stereolabs
4MP1080p HD, 30 FPSWVGA, 100 FPS
380mA / 5V170g
90◦, 60◦, 100◦
2 449.00
